Fast Source Identification Techniques For Digital Images

Source identification of digital image is to trace the source of digital image throughanalyzing and detecting the device noise within the digital image which arises from theimaging process of imaging device. With the explosive growth in the amount of media data,most current source camera identification methods are impractical because of low efficiencyof identification and unstable performance. So it has great significance to carry out thisresearch on how to improve the efficiency and robustness of source identification techniques.Besides, this research is also the key to determine whether those techniques can be wildlyapplied.This paper will combine the theoretical analysis and experimental verification for researchon fast source identification techniques of digital image. Based on simulating and analyzingthe typical algorithms, we propose some simpler, more efficient and robust algorithms on fastsource identification of digital image with the help of some special data structure, the idea offast algorithm and the method of pattern identification. The main contents and contributionsare described as follows.1. Review of the previous source identification algorithms of digital image anddiscuss the source identification algorithms based on imaging sensor pattern noiseand correlation detector. We set up a sizeable database with a number of pictures,which is used to simulate and analyze the current representative algorithms on fastsource identification of digital image. We point out the advantages anddisadvantages of those algorithms according to experimental results, which givethe research direction and content of our follow-up work.2. Aiming at improving the efficiency and robustness of source identificationtechniques, this paper proposes a fast algorithm for source camera identificationwhich uses sign information of large-magnitude components in camera sensorpattern noise. Most current source camera identification methods aretime-consuming and need large computational load, which could not meet thepractical needs. We design a simpler and more practical algorithm on fast sourceidentification of digital image with the help of some special data structure and theidea of fast algorithm.3. Aiming at the problem that most current source camera identification methods aresignificantly dependent on the training sample, this paper proposes a fastunsupervised clustering algorithm of unknown camera fingerprints. We design the fast unsupervised clustering algorithm according to the sign information betweencamera fingerprints and the idea of hierarchical clustering. With no a priorknowledge of the large database and no extra auxiliary images for train a classifier,the proposed algorithm can classify the camera fingerprints accordingly andestimate the number of sources camera.